The Neoplatonic Library

A citable library of restored public-domain editions of the Neoplatonic tradition — the texts, the whole scan each reading was decided from, and the repair log that records every editorial emendation.

Live at https://neoplatonic-library.org.

data/editions/<slug>/
    edition.json              bibliographic record, version, licence, citation, DOI
    versions/<semver>/
        source.txt            the served transcription, frozen at this version
        repairs.json          the repair log — every rule, with type, location,
                              before/after, rationale, witness and the leaf it
                              was read from
        meta.json             date, note, source hash, citation
        base.json             the edition's shared damage policy
    scans/                    the WHOLE scanned book, leaf by leaf
    import.json, scan.json, witnesses.json, derivs.json …
                              the source provenance records
data/graph/graph.json         the transmission / citation graph (nodes + edges)
data/site.json                site identity, licence, citation policy
build/                        the static site build (no server, no framework)
tools/                        the reader/extract/migration/verification tooling
design/library.css            the design system ("Reading Room")
design/DESIGN-SYSTEM.md       the design contract
docs/DATA-MODEL.md            the data model
elm/                          the reader client (Elm), compiled to tools/app.js

Each edition is a versioned, frozen object. source.txt never changes under a version; a correction ships as a new version and the old one keeps resolving.

/ · /texts · /texts/<slug> · /texts/<slug>/apparatus · /texts/<slug>/scans · /texts/<slug>/v/<semver> · /texts/<slug>/v/<semver>/apparatus · /graph · /search · /editions · /about · /errata.

The repair log is browsed as a leaf viewer on its own page, /texts/<slug>/apparatus/: the page images are the unit of provenance, so a reader pages through the scan and sees the readings decided from each leaf.

Everything the site serves is also plain data: /data/corpus.json, /data/graph.json, and /texts/<slug>/apparatus.json.

node build/build.mjs        # → site/dist/
./tools/deploy.sh           # rsync site/dist/ to the host

The build needs only Node — no pandoc, no framework. Verify with the probes in tools/ (apparatus-probe, route-probe, search-probe, scan-probe, leak-probe, exports-probe, version-probe, repair-id-probe, standalone-probe, heads-probe, library-smoke, library-extract-smoke, library-app-smoke).

Find as many copies as exist — scans, transcriptions, other printings of the same edition — measure their OCR, and build on the cleanest one. Hold the others as witnesses.

This is not optional polish. A book added from the first copy to hand was later found to be built on the weakest of four scans of the same 1816 print — 85% plausible tokens against 97% for the best. The reading view came out unreadable — "…{ ireames against the bertien 9 dat nee 0 Valet…" where a better copy reads "Indeed, that after the great incomprehensible cause of all, a divine multitude subsists…" — and a day of repair could not fix it, because repair cannot recover a bad base. The copy chosen also becomes the edition's scans, so it is a data decision as much as a textual one.

Check coverage, not just quality. In that case the two cleanest copies turned out to carry only the first volume; the cleanest copy of the whole work was a third scan. Score every candidate, then keep the ones that carry what the edition needs — the others are witnesses, and any part no clean copy carries has to be sourced from the best copy that does.

The test is cheap: sample a few hundred thousand characters from the middle of each candidate and score the share of word-shaped tokens. It separates the copies in seconds, and it should be run before anything is imported.

The transcriptions are public-domain works. The editorial work — repairs, page models, the apparatus, the notes — is released under CC BY 4.0. See LICENSE and LICENSE-editorial.

Cite the edition, the version, and the URL. Every edition page carries a formatted citation string. For example:

Proclus, The Elements of Theology, trans. Thomas Taylor (London, 1816). The Neoplatonic Library, version 1.0.0. https://neoplatonic-library.org/texts/proclus-elements-of-theology-taylor-1816/

The texts stand on their own; the essays at the blog are one reader's reading.